Can making 100 million court decisions searchable improve outcomes for people without lawyers? SFU computing science professor, Angel Chang, and Caseway begin research to find out.
Professor Chang is leading a planned research collaboration with Vancouver-based startup Caseway AI to examine a simple but high-impact question:
If court decisions are made fully searchable and usable by modern AI systems, will self-represented and marginalized people make better legal decisions?
Indexing a 100 million court decisions
The collaboration is being developed as a Mitacs-funded project. While the funding application is yet to be submitted, SFU and Caseway have already begun technical and research work due to the scale of the data and the urgency of access-to-justice challenges. Building and indexing a 100 million court decisions in Canada and the United States is a multi-year effort, and both teams agreed that early progress was critical.
At the center of the research is Caseway’s effort to publish and index more than 100 million Canadian and United States court decisions in a format that is searchable not only by humans, but by large language models such as ChatGPT and other AI systems. Each decision is structured, indexed, and made discoverable through modern search engines, with the goal of becoming a reliable, authoritative source that artificial intelligence systems can reference directly.
Angel Chang, professor in SFU’s School of Computing Science, emphasizes the importance of testing this rigorously:
“This research is not about replacing lawyers or automating legal advice. It’s about asking a careful, evidence-based question: if people without lawyers can access accurate, searchable court decisions through systems that use artificial intelligence, does that change how they understand their options and make early legal decisions? That’s what we want to measure.”
Hallucinations with LLMs
Today, most general-purpose AI tools lack access to real court decisions. As a result, they often rely on secondary sources such as blogs, forums, or summaries, which leads to hallucinations, missing context, and misleading legal information. Caseway’s approach is different: publish the primary source itself. When AI systems surface legal information, they can link back to the official court decision hosted on Caseway, allowing users to verify claims against the original judicial text.
The planned research will evaluate whether this shift in the information environment changes outcomes for people without lawyers. Rather than testing whether AI can give legal advice, the project focuses on whether access to accurate, searchable precedent improves understanding, confidence, and early-stage decision-making. For individuals going through the court system without a lawyer, even small improvements in comprehension of prior cases and legal standards may have meaningful downstream effects.
Retrieval system design and embedding experiments
Under Dr. Chang’s supervision, SFU students are already contributing to early prototyping work, including retrieval system design, embedding experiments, and evaluation of ranking quality. These efforts are laying the groundwork for future human-centered studies that would examine how people actually interact with search results and AI-assisted explanations when preparing for legal proceedings.
The work also fits into a broader academic and national context. Caseway is concurrently collaborating with researchers at the University of British Columbia on related legal AI research, reflecting growing interest across Canadian universities in domain-specific, responsible AI systems. Together, these projects explore how making primary legal sources machine-readable and publicly accessible could improve both AI accuracy and public understanding of the law.
Alistair Vigier, CEO of Caseway, says the project is about fixing the data foundations that AI systems rely on:
“Right now, most AI systems answer legal questions by pulling from Reddit threads, blogs, and second-hand commentary because the real court decisions simply aren’t accessible to them. Our goal is to change that by making official judicial decisions searchable at scale and usable by modern language models, so when artificial intelligence explains the law, it can point directly to the same sources judges rely on.”
Does better access to real court decisions lead to better outcomes?
A core motivation for the research is how today’s large language models actually learn legal information. General-purpose systems are trained on a mixture of licensed data, human-created data, and publicly available content. That public data heavily includes forums, discussion boards, and social platforms such as Reddit. This means that when users ask legal questions, models often surface explanations derived from informal commentary, anecdotes, or secondary interpretations rather than primary legal sources.
Multiple studies and disclosures have shown that Reddit is one of the most influential publicly available text sources shaping how LLMs reason and respond, particularly for question-and-answer style prompts. While this can be useful for everyday topics, it poses clear risks in legal contexts, where accuracy, jurisdiction, and precedent matter.
By publishing and indexing more than 100 million court decisions in machine-readable, search-engine-accessible formats, Caseway enables LLMs to reference official judicial texts directly and link users back to the original decisions.
The central hypothesis is not that AI should replace legal professionals, but that grounding AI outputs in real precedent rather than online commentary can materially improve accuracy, trust, and decision-making for self-represented individuals navigating the justice system.
Rather than asking whether AI should replace lawyers, the SFU–Caseway collaboration asks a narrower, testable question… Does better access to real court decisions lead to better outcomes? By making judicial decisions searchable at scale and usable by modern AI systems, the project aims to generate evidence that can inform future policy, research, and public-facing legal technology.
If successful, the work could help establish a new standard for how legal information is published and used by AI systems. It would be transparent, verifiable, and grounded in the same sources judges themselves rely on.